foo wsa:action (was WS-Addr issues)

Paco wrote:

> Why would someone need to shovel "foo" there? According to the current
> spec, you can either define your own value and attach it to the WSDL using
> the wsa:action attribute, or (provided you claim to support WS-Addressing)
> a value is derived for you based on the WSDL definition of the
> operation/message. No need for using "foo" anywhere.

within BT, and in the absence of a standard, we developed and deployed
our own addressing structure. This has no action field - we continue to 
dispatch a message to an 'operation' within an endpoint using the GED. 
In some cases there is no dispatching since the endpoint queue is
dedicated to perform a single action.  So i agree in either case a mandatory 
wsa:action wouldn't have "foo"  it would have something akin to the HTTP 
SOAPAction value.
 
However, i do see great architectural merit in using a header value to 
perform dispatching, not least we don't have to create a second envelope
for each message inside the body.  It seems that an optional wsa:action is 
primarily targeted at allowing less explicit dispatching mechanisms such 
as routing on the message contents or out-of-band data, not
least since it's currently difficult to describe such things. Also we
can't see how to access control such exchanges from a generalised 
message level proxy.  .  
 
i said earlier that i hadn't seen any arguments for the status quo, but 
having caught up with the 240+ messages sent to the list since wednesday, 
i've spotted several from DaveO, yourself and Gudge. I have to confess to 
feeling some very slight discomfort with some of this given it reminds me of 
the WSDL GED discussion only with the proponents reversed. It seems that 
some of the arguments used against prescribing the dispatching mechanism 
in WSDL could be used against prescribing them in addressing, not least 
that mandating wsa:action could be too restrictive  to be used in architectures
such as Savas and Jim's 'stuffHappens'. It is in such architectures that i see 
'foo' wsa:action values occuring, in the same way that they'll have to supply 
a null WSDL dispatching feature. 
 
Ultimately either a mandatory /or/ optional wsa:action would be 
acceptable. The only thing i'm looking for here is some way 
that takes the WG forward without risking the widespread adoption or
longevity of the spec, and that seems to be the status quo. 
 
Paul

Received on Monday, 8 November 2004 14:31:34 UTC